MIT Department of Electrical Engineering & Computer Science

DATE: THURSDAY, APRIL 13, 1995

TIME: TALK AT 4:15
Refreshments at 4:00

PLACE: NE43-518

Alignment by Maximization of Mutual Information

Paul Viola

Over the last 30 years the problems of image registration and recognition have proven more difficult than even the most pessimistic might have predicted. Progress has been hampered by the sheer complexity of the relationship between an image and an object, which involves the object's shape, surface properties, position, and illumination.

Changes in illumination can radically alter the intensity and shading of an image. Nevertheless, the human visual system can use shading both for recognition and image interpretation. I will present a metric for comparing objects and images that uses shading information, yet is explicitly insensitive to changes in illumination. This metric is unique in that it compares 3D object models directly to raw images. No pre-processing or edge detection is required. The metric has been rigorously derived from information theory.

Alignment is accomplished by adjusting the pose of an object until the metric distance between image and object is minimized. I will present a gradient descent alignment procedure based on stochastic approximation that has an efficient implementation. Experiments demonstrate this approach aligning a number of complex object models to real images.

HOST: Prof. Gerald Sussman